Description
We're looking for a full-stack scientist to pioneer quantitative research efforts at Udio. You will build at the intersection of research, engineering and product, bridging disciplines by drawing on huge, one-of-a-kind proprietary datasets of music, metadata and user interactions/feedback.
Design & own evaluation/optimization frameworks for frontier music models. Dive deep under the hood of our music generation systems, applying computational & human resources to understand model capabilities and identify areas for growth. Build optimization loops and apply your findings to our pretraining, post-training and inference systems as applicable.
Drive product & research roadmap. Own our data roadmap end-to-end, formulating research questions, exploring/linking/expanding data sources and conducting experiments at your discretion. Your work will span data mining, machine learning, causal inference, survey design and more, and your results will be critical for decision-making in product development, research investment and overall business direction.
Build stable infrastructure. Your work will reach far beyond the jupyter kernel, manifesting in robust integrations with our research & product tech stacks, potentially in performance-critical paths. You'll also build large-scale standalone data processing systems, allocating resources as needed to manage the data ecosystem.
Champion scientific rigor. As our first quantitative researcher, you'll cultivate a culture of scientific rigor across the company and deepen common understanding of models, users and data. You'll proactively identify opportunities, define metrics, share results, and build a rigorous foundation upon which to understand our highly subjective domain.
We're looking for someone with deep quantitative expertise, preferably a Ph.D. in statistics, mathematics, physics, or another quantitative discipline, or 5+ years' industry experience as a quantitative analyst / data scientist. Autonomy & ownership are key, as you'll thrive in greenfield research domains, undefined product categories and small, flat teams. Engineering chops are also important, as you'll need to translate your ideas into clear, production-ready code and collaborate in an active research codebase.